Face Embedding

Returns a numeric vector describing the main face in an image, for face matching in your own system. This is a utility, not a detector: it gives no real or fake verdict.

Slug	ID	Status	Tagged	Model version
`face-embedding`	`9`	Active	No	`v1`

What it does

The largest face in the image is located and converted into a 512-number vector. Vectors of the same person are close together, so you can compare two faces by the cosine similarity of their vectors. Vectors are L2-normalized, so the dot product gives the cosine similarity directly.

It does not check whether the face is real, live or swapped. Combine it with face-liveness-detection or ekyc-deepfake if you need that.

Request

Slot	Accepted `contentType`
`original`	`image/jpeg`, `image/png`

{
  "taskTypeId": "9",
  "inputs": [
    { "slotName": "original", "fileName": "portrait.jpg", "contentType": "image/jpeg", "sizeBytes": 262144 }
  ]
}

See Running a job for the full create, upload and finalize flow.

Limits

File size: up to the account upload limit, 10 MB by default.
The image must contain a detectable face. If none is found, the job fails and its credits are refunded.

Result

Field	Type	Description
`faceVector`	number[]	The face embedding: 512 numbers, L2-normalized
`dimension`	integer	Length of `faceVector`; always `512`
`modelVersion`	string	Version of the model that produced the result

{
  "faceVector": [0.0121, -0.0344, 0.0587, "…509 more values…"],
  "dimension": 512,
  "modelVersion": "v1"
}

Embeddings from different modelVersion values are not comparable. Re-generate your stored vectors when the version changes.

Output artifacts

`kind`	File	Description
`result`	`result.json`	The embedding vector and its dimension

Cost and availability

Fixed price per job, set by your plan. The exact amount is returned as cost when you create the job. See Quotas & Credits.
Available on the cloud API; your plan decides which task types you can use.